Journal of Geophysical Research: Biogeosciences

The global carbon cycle in the Canadian Earth system model (CanESM1): Preindustrial control simulation

Authors


Abstract

[1] The preindustrial carbon cycle is described for the Canadian Centre for Climate Modelling and Analysis Earth system model (CanESM1). The interhemispheric gradient of surface atmospheric CO2 concentration (xCO2) is reversed from the present day, with higher concentrations in the Southern Hemisphere, and southward interhemispheric transport by the ocean, estimated at 0.38 Pg C yr−1. The seasonal cycles of xCO2 and surface CO2 exchange are dominated by Northern Hemisphere terrestrial processes; the ocean contribution to CO2 flux is in phase with the larger terrestrial flux in the tropics and out of phase in the extratropics. Ocean processes dominate the relatively small Southern Hemisphere variability. Interannual variability of land carbon exchange is much larger than ocean exchange; both are comparable to results from previously published models with possibly larger variability in the terrestrial flux. Terrestrial net primary production (NPP) is determined largely by water availability at low latitudes, with temperature becoming more important at high latitudes. Temperature and moisture affect both NPP and heterotrophic respiration such that respiration effects tend to dampen the effect of fluctuations in NPP on CO2 exchange. Ocean CO2 flux variability is controlled by a variety of physical and biological processes with greater control by physical processes in the tropics and a larger biological contribution in the extratropics. Ocean CO2 flux is more strongly correlated with tropical sea surface temperature (SST) than terrestrial, but the variance associated with tropical SST is larger on land, in absolute terms, because of the much greater total variance of the land carbon flux. A novel hypothesis is advanced to explain how biological drawdown can cause recently upwelled water to be a net sink rather than source for atmospheric CO2. This process occurs over large areas of extratropical ocean and forms a natural sink for atmospheric CO2 that is potentially sensitive to both ocean acidification and anthropogenic perturbations of the aeolian iron flux.

1. Introduction

[2] As atmospheric CO2 concentration increases and climate warms, there is potential for significant feedbacks between climate and the terrestrial and ocean carbon pools. Whereas climate models initially inferred the effect of CO2 on climate by specifying an “equivalent radiative forcing” to represent all well-mixed greenhouse gases [e.g., Kattenberg et al., 1996], individual greenhouse gas concentrations are specified directly and radiative forcing is calculated internally in more recent studies. Relatively new are models with an interactive carbon cycle and freely varying atmospheric CO2. These have not been extensively used in the climate projections summarized by the Intergovernmental Panel on Climate Change (IPCC) [Meehl et al., 2007], but as the number of such models increases, they are likely to play a larger role in future IPCC assessments. Of the eleven models that participated in the recent Coupled Climate Carbon-Cycle Model Intercomparison Project (C4MIP), seven are global climate models with fully dynamical oceans and atmospheres, and the rest are Earth system models of intermediate complexity [Friedlingstein et al., 2006].

[3] The carbon cycle in the preindustrial control climate of these models has generally not been described in detail. Of the models in C4MIP, only the National Center for Atmospheric Research (NCAR) model has had its preindustrial control state described in detail in the literature [Doney et al., 2006]. Understanding the models' preindustrial carbon cycle is important, however, because their responses to CO2 emissions in future scenarios differ. In almost all cases, carbon cycle feedbacks exacerbate rather than moderate CO2-induced global warming, i.e., the overall feedback between climate change and the carbon cycle is positive, but the magnitude of this feedback differs substantially among models [Friedlingstein et al., 2006; Denman et al., 2007; Meehl et al., 2007]. The internal variability of the simulated climate system is also important in evaluating whether it will respond to changing external forcing in a realistic fashion. However, robust observation-based estimates of climate variability go back only a few decades, and the observational record of carbon cycle changes is shorter still. On longer time scales, only local proxy-based reconstructions are available, and their extrapolation to global climate must be interpreted cautiously. Models can help to bridge the gap between direct and proxy-based observations through their ability to provide long global time series of climate variables, which can assist in distinguishing between natural variability and anthropogenic effects in the observational record.

[4] We describe here the major characteristics of the preindustrial control run of the Canadian Centre for Climate Modelling and Analysis Earth system model (CanESM1), emphasizing the processes that determine the distribution of atmospheric CO2 and surface-atmosphere CO2 flux in space and time. The spatial patterns of variability of surface CO2 flux in the model control simulation can elucidate physical controls on surface CO2 exchange and help to explain the variability of the preindustrial climate and carbon cycle. The preindustrial carbon cycle was variable and in some cases quite large changes occur rapidly [e.g., MacFarling Meure et al., 2006]; models can help to elucidate the role of unforced variability in the genesis of these.

2. Methods

2.1. Model Description

2.1.1. Physical Climate Model

[5] The physical model employed is the Canadian Coupled General Circulation Model version 3.5 (CGCM3.5), with fully dynamic atmosphere, ocean, and sea ice. Previous versions of the coupled model (CGCM1 and CGCM2) are described by Flato et al. [2000] and Kim et al. [2002]. The third-generation atmosphere model (AGCM3) is described by Scinocca et al. [2008]. The ocean model, including the carbon cycle, is described by Zahariev et al. [2008]. The atmosphere model is run at T47 spectral resolution, a 96 × 48 surface grid. The 192 × 96 ocean model, with four grid cells beneath each atmosphere grid cell, has a longitude/latitude resolution of approximately 1.875°. Unlike previous model versions, flux adjustments are not used. Physical land surface characteristics are modeled by the Canadian LAnd Surface Scheme (CLASS) [Verseghy, 1991; Verseghy et al., 1993]. There are three soil layers, of thickness 0.1, 0.25, and 3.75 m, with prognostic liquid and frozen soil water. Brief descriptions of the land, ocean, and atmosphere carbon cycle components are given below.

2.1.2. Terrestrial Carbon Cycle

[6] The Canadian Terrestrial Ecosystem Model (CTEM) is coupled to CLASS, with specified fractional coverage of different plant functional types (PFTs) corresponding to year 1850 [Wang et al., 2006]. While proportions of different PFTs are specified for each grid cell, carbon pools are dynamic for each PFT. CLASS and CTEM use the 96 × 48 atmosphere grid and simulate fluxes of water, energy, and CO2. There are nine PFTs, each with separate pools for leaf, stem, and root carbon, as well as litter and soil carbon (Figure 1). Submodels of specific terrestrial ecosystem processes are described by Arora [2003] and Arora and Boer [2003, 2005]. Fire and dynamic vegetation modules were not used in this simulation. A river-routing scheme [Arora and Boer, 1999] is used, but runoff transports no carbon or other chemical species.

Figure 1.

Schematic of the Canadian Terrestrial Ecosystem Model (CTEM). Root and stem allocations indicate that some fraction of gross primary production is allocated to these pools, with the remainder going to leaf production. Leaf, stem, and root respiration each contain a maintenance component and a growth rate-dependent component. All other arrows including maintenance respiration (except gross primary production) represent fluxes based on specified turnover rates that vary with temperature and soil moisture.

[7] Photosynthesis in CTEM, based on the studies by Farquhar et al. [1980] and Collatz et al. [1991, 1992], uses a single-leaf approach that couples photosynthesis and canopy conductance based on vapor pressure deficit [Leuning, 1995]. Photosynthesis operates at the atmosphere model time step of 20 minutes. All other submodules of CTEM operate at a daily time step, at which carbon fixed by photosynthesis throughout the day is allocated to leaves, stem, and root based on light, root water, and leaf phenological status. The phenology submodule uses a carbon-gain approach: leaf onset is initiated when photosynthetic gains exceed respiratory losses of virtual leaves [Arora and Boer, 2005]. Deciduous PFTs lose their leaves under unfavorable environmental conditions of insufficient irradiance, low temperature, or low soil moisture; there is a low-latitude drought-deciduous PFT as well as the midlatitude seasonal leaf-out [Arora and Boer, 2005]. Allocation of carbon to the three vegetation components (leaves, stem, and root), and losses to respiration and litterfall, determine the aboveground biomass that in turn determines the structural vegetation attributes used in the energy and water balance calculations [Arora and Boer, 2005]. Leaf area index is a function of leaf biomass, and vegetation height is a function of aboveground biomass. In addition, root biomass determines an evolving root depth distribution profile [Arora and Boer, 2003] that also affects evapotranspiration rate. CTEM does not explicitly model nutrient limitation; an empirical parameterization of nitrogen limitation is used in simulations with anthropogenic emissions [Arora et al., 2009], but this effect is not important in the preindustrial control simulation.

[8] Gross primary production (GPP) is the total uptake of carbon by the vegetation; net primary production (NPP) is GPP less the respiration of the three vegetation fractions (leaf, stem, and root) (Figure 1). Heterotrophic respiration (RH) is respiration of soil and litter carbon (Figure 1). Net ecosystem production (NEP) is the difference between NPP and RH (negative when RH > NPP). NEP is equivalent to net terrestrial C uptake (equivalent to the net surface CO2 flux over land but positive downward).

2.1.3. Ocean Carbon Cycle

[9] The Canadian Model of Ocean Carbon (CMOC) is a global ocean general circulation model (OGCM) with embedded carbon chemistry and biology modules [Zahariev et al., 2008]. Carbon chemistry and gas exchange follow the OCMIP II protocols [Najjar and Orr, 1998]. The ecosystem model is a simple NPZD model (Figure 2) with nitrogen as the primary currency. The carbon cycle is tied to the ecosystem through a fixed Redfield ratio of 6.625 and parameterizations for calcium carbonate precipitation and dissolution as described below. New nitrogen is introduced via dinitrogen fixation in tropical and subtropical oceans as described in Zahariev et al. [2008]; denitrification is parameterized very simply so that the total fixed N pool is conserved. Alkalinity is also conserved globally: losses to calcium carbonate burial at the sea floor are returned at the surface as a crude parameterization of river inputs of alkalinity.

Figure 2.

Schematic of the ocean ecosystem model (CMOC). Ecosystem compartments are inorganic nitrogen (N), phytoplankton (P), zooplankton (Z), and detritus (D). Chlorophyll (Chl) is a semiprognostic quantity derived from phytoplankton by a light-dependent chlorophyll-to-nitrogen ratio but carried as a separate tracer. Sedimentation of organic carbon is derived from detritus concentration, with a sinking velocity of 10 m d−1; sedimentation of inorganic carbon is calculated from the organic flux at the base of the euphotic zone via a temperature-dependent rain ratio.

[10] A single nutrient compartment represents all dissolved N species. Phytoplankton growth is limited by light, nutrient (N), and iron. Iron limitation is parameterized by a geographic “mask” based on the present-day distribution of “excess” surface nitrate as described by Zahariev et al. [2008]. Zooplankton grazing follows a quadratic parameterization that limits grazing pressure at very low prey concentrations. Detritus sinks at a constant speed and remineralization is temperature-dependent so that, on average, the remineralization length scale will increase with increasing depth. The vertical flux of organic carbon is determined by the flux of detrital N via the Redfield ratio; the flux of particulate inorganic carbon is calculated from the organic flux at the base of the euphotic zone using a temperature-dependent “rain ratio” [Zahariev et al., 2008]. Dissolution of particulate inorganic carbon is parameterized as exponential decay of the vertical flux, i.e., transport to depth is local and instantaneous. Complete conservation equations for ecosystem compartments N, P, Z, and D, and dissolved inorganic carbon (DIC) and alkalinity, are given by Zahariev et al. [2008].

2.1.4. Atmospheric CO2 and Other Greenhouse Gases

[11] There are no sources or sinks for CO2 within the atmosphere. CO2 is carried as a 3-D advected tracer and treated as such in the radiation calculations. Other major greenhouse gases such as CH4 and N2O are assumed to be spatially uniform and are treated individually in the radiation code, with concentrations fixed at 1850 levels (792 and 276 ppb for CH4 and N2O, respectively). The solar constant is 1365 W m−2. Sulfate aerosols were specified as aerosol optical depth (AOD) and prescribed values for 1850 were used. AOD is used to increase the surface albedo to mimic cooling caused by aerosol scattering of incoming solar radiation [Reader and Boer, 1998].

2.1.5. Spinup Procedure

[12] The ocean and terrestrial carbon models were first spun up in a decoupled mode with a fixed atmospheric CO2 concentration of 288 ppm, corresponding to base year 1850. The terrestrial carbon model was forced with an observation-based compilation of atmospheric forcing for the years 1979–1999 at 6 hourly resolution [Dirmeyer and Tan, 2001] in a repeating cycle for >1000 years. The ocean was forced offline with a repeating climatological annual cycle of CGCM3.5 wind stress and heat and freshwater fluxes for 6000 years [Zahariev et al., 2008]. The run from which this forcing was derived did not include sulfate aerosol, resulting in a slight drift to a cooler climate and lower atmospheric CO2 following coupling; the mean surface atmospheric CO2 concentration (xCO2) dropped from 288 to 285 ppm (Figure 3 and Table 1). Even after 600 years, a slight drift is present in the integral land and ocean carbon pools, but these largely offset each other (Figure 3). The mean drift in the total atmospheric CO2 is 0.001 Pg C yr−1 for years 601–1600, against a total of 604 Pg, or ∼0.2% per thousand years. The drift in total terrestrial carbon storage over 1000 years is around 4 Pg C (Figure 3), similar to the atmosphere in percentage terms, while in the ocean, the relative drift is much smaller due to its larger total carbon content. Data presented here represent annual means for years 601–1600, unless otherwise stated; year 601 is labeled Year 1 in Figure 3 and all subsequent figures and tables. The results presented are not sensitive to the particular time periods chosen.

Figure 3.

Time series of global mean (〈X〉) surface temperature and surface atmospheric CO2 concentration (xCO2), and total land, ocean, and atmosphere carbon pool anomalies (annual means for years 601–1600) relative to long-term mean (〈equation image〉). 10 Pg C was added (subtracted) to the land (ocean) carbon pool for easier visualization.

Table 1. Climatological Mean and Standard Deviation of Globally Annually Averaged Surface Temperature and CO2 Concentration and Total Ocean and Land CO2 Flux, for 1000 Years of the Control Runa
 MeanStandard Deviation
  • a

    See section 2.2 for definitions. CO2 fluxes are positive upward. Standard deviations of CO2 fluxes have linear trends removed (see Figure 3).

Temperature (°C)13.480.061
xCO2 (ppm)285.10.65
Land-atmosphere CO2 flux (Pg C yr−1)0.00960.88
Ocean-atmosphere CO2 flux (Pg C yr−1)−0.00920.17

2.2. Notation and Statistics

[13] The vertically integrated carbon budget for preindustrial conditions and with no anthropogenic emissions may be written as

equation image
equation image
equation image

for the three carbon pools, the atmosphere, land and ocean. Here HA, HL, and HO are the atmosphere, land, and ocean carbon stores, respectively; HA and HO are the horizontal carbon transport vectors for the atmosphere and ocean, respectively; and FA, FL, and FO are the fluxes of carbon between components at the surface (positive upward). Combining the three equations gives the budget for the entire system as

equation image

where H = HA + HL + HO and FA − (FL + FO) = 0.

[14] We express climatological statistics of terms related to the carbon budget in the form

equation image

where 〈X〉 represents a spatial average (often global) of X and X+ the deviation from this global average (for 3-D fields, the spatial average may include the vertical as well as the horizontal dimension). equation image is the time average and X′ the deviation from the time average. 〈equation image〉 is the global spatial and temporal average, and equation image+ is the time-averaged spatial deviation from 〈equation image〉. X* = equation image+ + X′ is the deviation from the area and time average.

2.2.1. Correlation and Regression Analysis

[15] The correlation coefficient for point-by-point correlation maps is

equation image

and for the total space time correlation it is

equation image

where X* = equation image+ + X′ (see equation 3). A linearized decomposition of the local terrestrial CO2 flux anomaly relative to the global time-space mean can be written as

equation image

where F represents CO2 flux and T, S, and Q represent temperature, soil moisture, and surface downwelling solar irradiance at the surface. The regression coefficients aX represent “loading factors” that express the amplitude of the change in CO2 flux anomaly associated with fluctuations in each independent variable X, which can be estimated by multiple linear regression analysis. These regression coefficients vary independently of the correlation coefficients in part because of covariance among the independent variables and in part because σ is evaluated over the full time-space domain. Normalization by σ is necessary for direct comparison of regression coefficients that would otherwise implicitly have different units. Multiple linear regression analysis can potentially produce misleading results because a statistically “optimal” solution with, e.g., a1a2 may be only marginally better, in a least-squares sense, than an alternative solution with a2a1. Such results cannot be definitively excluded; they can only be discounted by a combination of a priori physical reasoning and a posteriori examination of the parameter space and parameter estimation errors. Nonetheless, such an analysis provides additional information about the amplitude of the response of FL to environmental factors that correlation analysis alone does not provide.

2.2.2. Atmospheric Transport

[16] The vertically integrated and time-averaged CO2 budget equation is

equation image

where C is the mass mixing ratio of CO2, V is the horizontal velocity vector, FA is the flux of CO2 into the column at the surface, dp/g = ρdz is the element of mass in the pressure coordinate system, and the overbar indicates the long-term mean. For the control simulation, the rate of change term is small so that to a good approximation

equation image

which equates the flux of CO2 into the column at the surface with the divergence of the horizontal flux vector H = ∫equation image. The flux vector may be decomposed into its rotational and divergent components H = HR + HD, where ∇ · HR = 0, so that it is the divergent flux component HD = ∇χ, where χ is a potential function, which is connected to the surface CO2 flux in (7) [Boer and Sargent, 1985].

2.2.3. Monte Carlo Analysis of Power Spectra

[17] The significance of deviations of power spectra from a null hypothesis of “white noise” was tested using a Monte Carlo method where 1000 time series of Gaussian random numbers with the same mean and variance as the data were generated (after confirming that the data distribution was approximately Gaussian). These were transformed in the same way as the data in order to define upper and lower 95% confidence intervals for the amount of energy in a particular frequency band. The range of frequencies in which significant (P < 0.05) deviations occurred was then tested for significant deviations from a random sample of the N frequencies resolved by 1000 trials using uniform random integers ≤N.

3. Results

3.1. Variability of the Global Mean State

[18] Time series of the global mean surface temperature 〈T〉 and xCO2 〈C〉, and total land, ocean, and atmosphere carbon pool anomalies (〈HL〉′, 〈HO〉′, 〈HA〉′) are shown in Figure 3. Temperature and xCO2 are stable at around 13.5°C and 285 ppm, with ranges of about 0.4 K and 4 ppm, respectively (Figure 3). These ranges are very similar to those of the NCAR model [Doney et al., 2006]; however, that model shows relatively large centennial-scale excursions in xCO2, whereas ours is less variable at that time scale. Some abrupt drops like that observed by Macfarling Meure et al. [2006] are present, but the magnitude is smaller by about a factor of 2. Because atmospheric CO2 is relatively well mixed, the global atmosphere burden is closely related to xCO2 (r = 0.94). Interannual to interdecadal variability in the atmospheric CO2 burden is closely related to variability of land carbon storage: increases in terrestrial storage result in similar decreases in the atmosphere burden but may be damped by variations in ocean storage (e.g., around year 460). The total land (〈HL〉′) and ocean (〈HO〉′) carbon pool anomalies (shown here with offsets of +10 and −10 Pg for clarity) each drift slightly (<0.01 Pg C yr−1), but in opposite directions (Figure 3 and Table 1), so the net impact on the atmosphere burden is small.

[19] Means 〈equation image〉 and standard deviations σequation imageXequation image for global annual mean quantities are shown in Table 1. Global mean temperature is consistent with observations and other climate model values; the range of temperatures (Figure 3) is similar to, or slightly smaller than, historical reconstructions of Northern Hemisphere mean temperature [Jansen et al., 2007]. xCO2 has a standard deviation of less than 1 ppm. Land-atmosphere and ocean-atmosphere fluxes have means near zero as expected; the slightly negative and positive values, respectively, reflect a very slow transfer of carbon from land to ocean due to climate drift. (CO2 fluxes are always presented as positive upward, i.e., a negative value represents a loss from the atmosphere to the ocean or terrestrial biosphere.) The standard deviation of the globally averaged land-atmosphere flux (σequation imageimageequation image) is greater than that of the ocean-atmosphere flux (σequation imageimageequation image) by about a factor of 5.

[20] Power spectra of the net land-atmosphere and ocean-atmosphere CO2 fluxes (〈FL〉, 〈FO〉) are shown in Figure 4. The spectra are approximately white, i.e., there are no frequency bands that stand out as containing more energy than the others. The ocean spectrum is slightly “redder” than for land, especially at the higher frequencies (e.g., >0.03 yr−1), where ocean variability increases toward lower frequencies. Land variability is generally greater across the range of frequencies. The spectra are similar to those of Doney et al. [2006] but with slightly more variability in the land-atmosphere flux and a less “red” ocean at low frequencies. The model lacks the sort of centennial-scale excursions observed by Doney et al. [2006], but it is not known whether such variability is realistic. The power spectra confirm that at subannual frequencies, the total land-atmosphere flux is considerably more variable than the ocean-atmosphere flux (see also Table 1). This is consistent with observation-based estimates and other model studies [e.g., Zeng et al., 2005; Doney et al., 2006; Denman et al., 2007].

Figure 4.

Power spectra for globally integrated land-atmosphere (FL, dashed line) and ocean-atmosphere (FO, solid line) CO2 flux, from 1000 years of annual mean output.

[21] There appear to be statistically significant deviations from the null hypothesis of white noise (variance in more than 5% of frequency bands exceeds the upper 95% confidence limit estimated with a Monte Carlo method). For the ocean, the variance exceeds the upper 95% confidence limit at 13.2% of frequencies resolved, while only 3.6% are below the lower confidence limit. For the land flux, the upper 95% confidence limit is exceeded at 6.0% of frequencies and the lower at 7.2%. The apparent concentration of positive deviations from white noise in the ocean-atmosphere flux spectrum in the interannual-to-interdecadal range is difficult to interpret, as this subrange occupies a fairly large fraction (∼80%) of the frequencies resolved. However, a random sampling of these frequencies includes no values outside this range in less than 1% of trials. The land spectrum is below the 95% confidence limit at all frequencies below 0.01 yr−1, and the probability of this occurring by chance is less than 0.001.

[22] Temporal autocorrelation of global mean or integral pools and fluxes is shown in Table 2. Autocorrelation of global mean surface xCO2 (〈C〉) and total terrestrial carbon storage (〈HL〉) exceeds 0.8 for a lag of 1 year and remains significant for lags up to 10 years (Table 2). The ocean carbon storage anomaly (〈HO〉′) has the longest “memory” of the quantities examined, with autocorrelation exceeding 0.8 for lags up to 10 years (Table 2). Terrestrial CO2 flux (〈FL〉) shows little autocorrelation at any lag; the ocean flux (〈FO〉) also shows little autocorrelation at lags greater than 1 year (Table 2). The autocorrelation of the mean surface xCO2, therefore, principally reflects that of the total atmospheric burden, because of the inertia of the ocean and land pools, rather than any modulation of surface exchange processes by climate variability. However, there is some evidence for such a modulation of the ocean CO2 flux on an interannual time scale, as both surface temperature and ocean CO2 flux show an autocorrelation of ∼0.4 at lag 1 year (Table 2).

Table 2. Autocorrelation in Time of Global (Land, Ocean) Annual Mean Screen Temperature, Mean Surface CO2 Concentration, Total CO2 Flux, and Total C Storage Anomalya
Lag (y)〈T〉〈C〉〈FA〈FO〈FL〈HO〉′〈HL〉′
  • a

    A, atmosphere; L, land; O, ocean. Values less than 0.10 (absolute value) are considered not significant and are not shown; this significance cutoff is itself a function of autocorrelation and is somewhat conservative. Linear trends were removed from the land and ocean total C storage (see Figure 3).

10.440.85−0.100.39−0.100.980.88
20.220.66−0.14 −0.120.960.77
30.140.55−0.11−0.11−0.110.930.70
4 0.49 −0.17 0.910.65
5 0.45 −0.11 0.890.62
6 0.41   0.870.58
7 0.37   0.860.55
8 0.34   0.850.52
9 0.32   0.840.50
10 0.31   0.820.49

3.2. Global Spatial Patterns

[23] Distribution of the long-term means (equation image) and standard deviations (σequation image) of annually averaged near surface temperature and precipitation are shown in Figure 5 (screen temperature is the estimated temperature 2 m above the surface). Contemporary observed fields (ERA40 for 1958–2001 for temperature and GPCP for 1979–2003 for precipitation) are used for comparison as there are no direct observations of the preindustrial climate. Differences from the observed temperature are concentrated in the high latitudes and over mountainous regions because of the model's coarse discretization of the topography (Figure 5e). Temperature differences in the northern high latitudes are exacerbated by anthropogenic warming in the observational record but are too large to attribute to this alone. Temperatures over the ocean are biased warm over a large area of the southern midlatitudes (Figure 5e). For precipitation, differences from the observed field can be large in the deep tropics due to shifts in the latitudinal positions of the rain bands. Modeled precipitation is low in the Pacific Intertropical Convergence Zone and over parts of Amazonia and too high in the equatorial zone (Figure 5f). Temperature variability (standard deviation of annual mean values) is greatest over land and at high latitudes. Temperature variability is underestimated in the tropical ocean (not shown), where interannual variability is large in the real world [e.g., Philander, 1990]. Precipitation variability is highest in the tropics, at least in part because the mean precipitation is higher (Figures 5b and 5d). Precipitation variability has been shown to compare reasonably well with observation-based estimates in earlier but similar versions of the physical climate model [Scinocca and McFarlane, 2004; Arora and Boer, 2006].

Figure 5.

Global distribution of time mean (equation image) and standard deviation (σequation image) for 1000 years of annual mean screen temperature and precipitation and deviations from present day observations. Observed temperatures are from 40 years of ERA reanalysis and precipitation from 25 years of GPCP data. Note that in Figure 5c the color scale is discontinuous, with a single color for values exceeding 1.8°C.

[24] Simulated xCO2 anomaly (equation image+) and surface CO2 flux (equation image) are shown in Figure 6. Surface xCO2 varies over a range of around 4 ppm, with slightly higher values in the Southern Hemisphere (Figure 6a), so that the north–south gradient is reversed from the present day [Taylor and Orr, 2000]. Local maxima of equation image+ tend to occur over regions of net outgassing (equation image > 0) from the oceans, e.g., the tropics and high southern latitudes, and minima over ocean regions of net uptake (equation image < 0) such as the subantarctic and the poleward ends of the Northern Hemisphere western boundary currents (Figure 6b). The ocean largely controls the temporal-mean spatial pattern of xCO2 because lateral transport of carbon within the ocean supports regions of consistent outgassing or uptake (Figure 6b). On land, the temporal-mean net fluxes are uniform and near zero, since there is no net lateral transport and respiration balances photosynthesis locally (Figure 6b). Variability of both xCO2 and CO2 flux is larger over land than over the oceans and generally larger at lower latitudes (Figures 6c and 6d), consistent with other studies [e.g., Zeng et al., 2005].

Figure 6.

Global distribution of time mean (equation image) and standard deviation (σequation image) for 1000 years of annual mean CO2 mixing ratio (xCO2) anomaly (equation image+) in the lowest layer of the atmosphere and surface CO2 flux (equation image, positive upward). Figures 6c and 6d use bilinear scales to emphasize differences in the upper part of the range.

[25] The vertically integrated horizontal transport of CO2 in the atmosphere (HA) can be obtained from the surface flux as described in section 2.2.2. This net transport, shown in Figure 7, is divergent in the eastern tropical Pacific and the high-latitude Southern Ocean and convergent over the Kuroshio extension and the subantarctic Southern Ocean, areas of net ocean uptake as noted above. Transport over the continents is generally small, although there is some net northward transport over North America (transport over land is nondivergent because the time-mean net source or sink on land is zero or very small). The meridional component of the vectors nearest the equator is either near zero or oriented northward, consistent with a small net transport from the Southern Hemisphere to the northern, counterbalanced by a southward flux within the ocean (see section 4). There is little net transport from the southern subtropics into the sink regions of the midlatitude Southern Ocean; most of the net uptake is from the high-latitude Southern Ocean source (see section 4).

Figure 7.

Vertically integrated horizontal transport of CO2 in the atmosphere (equation imageA) calculated from surface sources and sinks (equation image). Color scale shows the surface CO2 flux (molC m−2 yr−1), contours show the potential function (molC s−1 × 104, see text), and vectors show the net transport (molC m−1 s−1 × 102).

[26] The vertical distribution of CO2 in the atmosphere is shown in Figure 8 as the zonal mean anomaly relative to global annual mean (i.e., deviation of zonal mean equation image+ from 〈equation image〉). The annual mean is influenced by the strong source in the high-latitude Southern Ocean and a minimum in the high-latitude Northern Hemisphere associated with terrestrial uptake in northern summer (see below section 3.3). Despite uptake by the midlatitude Southern Ocean, the anomaly remains positive over almost the entire Southern Hemisphere atmosphere with the zero contour situated near the equator. Plots for January and July (Figure 8, middle and bottom) show the importance of tropical convection and seasonal Northern Hemisphere land sources and sinks in maintaining the global 3-D distribution. In the tropics, anomalies generated at the surface are transported rapidly to the tropopause and advected poleward, forming maxima or minima at heights of 200–500 hPa (Figure 8); the sign of these anomalies is out of phase with the midlatitude to high-latitude Northern Hemisphere land source/sink (see below section 3.3). The high-latitude Northern Hemisphere minimum is strongest in the midtroposphere in the annual mean, but this is largely due to temporal averaging, as in summer and winter the largest anomalies in the midlatitude and high-latitude Northern Hemisphere are at the surface (Figure 8).

Figure 8.

Vertical and latitudinal distribution of zonal mean atmospheric CO2 mixing ratio (ppm) anomaly (equation image+) relative to overall space-time mean 〈equation image〉. Annual, January, and July means calculated for years 400–450.

3.3. Seasonal Cycle

[27] The climatological seasonal cycles of surface xCO2 (Figure 9a) and CO2 flux (Figure 9b) are larger in the Northern than in the Southern Hemisphere because of the asymmetric distribution of land mass between the hemispheres. The terrestrial biosphere takes up CO2 from the atmosphere in the summer hemisphere due to an excess of photosynthesis over respiration, while CO2 is released in the winter hemisphere due to excess respiration (Figure 9d). Although ocean photosynthesis is seasonal and of comparable magnitude to terrestrial net primary production (NPP), effects of seasonal changes in ocean biological production on CO2 flux are largely offset by temperature-dependent changes in CO2 solubility, i.e., the largest biological drawdown of DIC occurs in summer when higher SST tends to increase pCO2 [e.g., McKinley et al., 2006]. The seasonal amplitude of surface CO2 and CO2 flux is small in the Southern Hemisphere and controlled largely by ocean processes (Figures 9a–9c); the largest seasonal amplitude is in the Northern Hemisphere and is primarily a function of terrestrial exchange (Figures 9a, 9b, and 9d). In the midlatitude to high-latitude Northern Hemisphere, the ocean and terrestrial contributions are out of phase, so the ocean damps the terrestrially controlled seasonal cycle slightly (Figures 9b and 9c). Ocean uptake is maximal in the winter hemisphere due to surface cooling and convective mixing [McKinley et al., 2006; Zahariev et al., 2008]. Outgassing due to warming of the sea surface in summer is limited by enhanced biological uptake [McKinley et al., 2006]; offline CMOC simulations show limited periods of summer outgassing in both the North Atlantic and North Pacific [Zahariev et al., 2008]. In the Southern Ocean, outgassing associated with equatorward Ekman transport is maximal in spring and fall (Figure 9c), consistent with the “semiannual oscillation” described by van Loon [1967].

Figure 9.

Seasonal cycle of zonal mean (a) CO2 mixing ratio (ppm) in the lowest layer of the atmosphere, (b) net surface-atmosphere CO2 flux (FL + FO), (c) ocean-atmosphere CO2 flux (FO), (d) land-atmosphere CO2 flux (FL) (all in mol m−2 yr−1), as a function of latitude. Climatology of years 400–450: one climatological year is repeated twice. Common color scales are used for total and land fluxes.

[28] Figure 10 shows the seasonal cycle of CO2 and CO2 flux in the low latitudes. Ocean-atmosphere exchange is approximately in phase with the land-atmosphere exchange, at least in the Southern Hemisphere (Figures 10c and 10d), and reinforces the terrestrial contribution to the seasonal cycle of atmospheric CO2 (Figure 10a). In the tropics, land uptake (Figure 10d) follows the seasonal cycle of rainfall. Precipitation occurs primarily in the summer hemisphere [Huffman et al., 1997], as does net terrestrial carbon uptake (Figure 10d). Tropical ocean outgassing is maximal in boreal summer, particularly in the Southern Hemisphere (Figure 10c). In the equatorial oceans, mean outgassing is strongest in the Southern Hemisphere (Figure 7) and the strongest upwelling favorable winds occur in boreal summer/fall [Philander and Chao, 1991]. Subtropical ocean uptake is largely temperature controlled and occurs primarily in the winter hemisphere (Figures 9c and 10c).

Figure 10.

As Figure 9 but for latitudes ≤30° only.

[29] The seasonal amplitudes of modeled temperature and precipitation are similar to observed values (Table 3). Observations of the seasonal cycle of atmospheric CO2 concentration are not available for the preindustrial period; model-data comparisons for the late 20th century show that the modeled seasonal cycle of global mean atmospheric CO2 concentration compares favorably with observations, with the model amplitude about 10% larger [Arora et al., 2009]. Qualitatively, the pattern shown in Figures 9a and 10a resembles the north–south gradient in the amplitude of the seasonal cycle in contemporary observations, which is not strongly affected by anthropogenic emissions [Masarie and Tans, 1995; Erickson et al., 2008].

Table 3. Seasonal Ranges for Surface Temperature and Precipitation, Compared with Observations for the Period 1980–2000a
 ModelObservedCMAP
  • a

    Seasonal ranges are the difference between the highest and lowest monthly values in the climatological seasonal cycle for each region. Model values are calculated for a climatology of years 400–450. Precipitation observations are from GPCP [Huffman et al., 1997]; alternate CMAP data set [Xie and Arkin, 1996] provided for comparison. Observed temperatures are from ERA40 reanalysis.

Temperature
   Northern Hemisphere13.812.8 
   Southern Hemisphere6.285.87 
   Tropics (30°S to 30°N)1.181.24 
Precipitation
   Northern Hemisphere1.61.31.7
   Southern Hemisphere1.61.31.5
   Tropics (30°S to 30°N)0.100.130.18

3.4. Controls on Interannual Variability

[30] As noted above, interannual variability of temperature, xCO2, and CO2 flux is greater over land than over the oceans. Temperature variability (σequation image) is greatest at high latitudes while xCO2 and CO2 flux variability (σequation image, σequation image) are generally larger at low latitudes (Figures 5 and 6). Environmental factors controlling the terrestrial carbon flux include temperature, precipitation, soil moisture, and solar radiation. Over the ocean, CO2 flux is controlled by surface temperature and salinity, wind speed, mixed layer depth, and biological uptake (storage as organic carbon and/or transport to depth).

[31] Figure 11 shows the correlation coefficient of annual mean CO2 flux (positive upward) with surface temperature, precipitation, and soil water content (point-by-point correlation, see equation 4a). Over land, the correlation with temperature is generally positive (71% of total land grid points), especially in the tropics, and negative over some high-latitude or high-altitude cold regions, notably eastern Canada, Alaska, eastern Siberia, and Tibet (Figure 11a). This implies a tendency toward reduced terrestrial carbon storage in the tropics and increased storage at high latitudes in a warmer climate. Correlation of terrestrial CO2 flux with both precipitation and soil moisture is generally negative (precipitation enhances land uptake) except at high latitudes (Figures 11b and 11c). These patterns are robust to low-pass filtering of the annual data (not shown) and are therefore indicative of environmental controls on decadal or longer as well as interannual time scales.

Figure 11.

Correlation coefficient (rxy) of surface CO2 flux (positive upward) with surface temperature, precipitation, and total soil water. Over land, CO2 flux is equal to NEP × −1.

[32] Because temperature, precipitation, soil moisture, and solar irradiance are all interrelated, both in the physical climate system and through ecosystem processes, separating their influence on ecosystem production and CO2 flux is not straightforward. For example, there may be a positive correlation between irradiance and heterotrophic respiration simply because increased primary production leads to increased production of litter (as well as covariance of irradiance and temperature). Table 4 shows the correlation coefficients across both space and time (rxyt, see equation 4b) of CO2 flux with environmental factors, and the regression coefficients that represent the amplitude of the variability in CO2 flux associated with each of these environmental factors (see section 2.2.1).

Table 4. Correlation and Regression Coefficients (“Loading Factors”) for Terrestrial Ecosystem Production Versus Mean Temperature, Soil Moisture, and Solar Irradiance in High- and Low-Latitude Ecosystemsa
  Correlation CoefficientLoading Factor
TSQTSQ
  • a

    Based on 200 years of monthly data, calculated over both time and space (equations (3)(5)), for mean seasonal cycle and for anomalies relative to mean seasonal cycle. Loading factors (absolute value) are shown only where ∣r∣ > 0.25. T, temperature; S, soil moisture; Q, solar irradiance; GPP, gross primary production; NPP, net primary production; NEP, net ecosystem production; RH, heterotrophic (soil and litter) respiration. NEP is equivalent to net terrestrial CO2 uptake.

0°–20°GPP−0.100.32−0.17 27.5 
SeasonalNPP−0.260.51−0.3914.716.42.4
 NEP−0.180.11−0.16   
 RH−0.200.53−0.37 5.40.6
        
0°–20°GPP−0.150.290.13 7.7 
AnomalyNPP−0.230.320.10 5.7 
 NEP−0.340.33−0.0117.45.3 
 RH0.53−0.200.461.4 0.4
        
0°–40°GPP0.360.48−0.0523.731.9 
SeasonalNPP0.310.56−0.139.218.1 
 NEP0.010.07−0.03   
 RH0.330.61−0.163.25.9 
        
0°–40°GPP−0.090.300.08 7.9 
AnomalyNPP−0.160.330.05 5.8 
 NEP−0.270.32−0.047.25.3 
 RH0.45−0.110.400.8 0.4
        
40°–90°GPP0.560.240.4013.2 1.9
SeasonalNPP0.550.210.418.3 0.8
 NEP0.370.030.362.4 1.5
 RH0.510.300.301.90.30.9
        
40°–90°GPP0.080.240.00   
AnomalyNPP0.030.24−0.04   
 NEP−0.090.16−0.13   
 RH0.310.130.240.2  

[33] In the tropics (0°–20°), both gross and net terrestrial primary production are negatively correlated with temperature (Table 4). Net ecosystem production (NEP), which is equivalent to net surface CO2 uptake or outgassing, is generally weakly correlated with physical environment variables (Table 4). Changes in temperature and precipitation can alter both NPP and heterotrophic respiration, with these effects offsetting each other so that the effect on NEP is small [Reichstein et al., 2007]. The relative importance of temperature control of both NPP and heterotrophic respiration increases with increasing latitude, which translates into an effect on NEP over the seasonal cycle (Table 4). The correlation coefficients shown in Table 4 describe general physical controls on plant growth, but because they include both spatial and temporal variation, they can mask relationships within specific regions (Figure 11) because different processes are operative in different climatic zones. In Amazonia and southeastern Africa, for example, there are reasonably consistent relationships: greater uptake of CO2 during wetter periods (negative correlation with precipitation and soil moisture) and outgassing during warmer ones (positive correlation with temperature) (Figure 11). Nonetheless, it is clear that heterotrophic respiration is controlled by the same factors that determine NPP (Table 4), as well as being strongly influenced by NPP itself.

[34] The (pointwise) correlation of ocean CO2 flux with temperature is strongly negative in the eastern tropical Pacific Ocean and generally positive in the subtropics (Figure 11a). Over the extratropical oceans, CO2 solubility (as a function of temperature and salinity) exerts an important direct control on CO2 flux, whereas in the tropical oceans, ocean dynamical processes largely control the interannual variability of CO2 flux [Christian et al., 2008]. CO2 flux is correlated with precipitation over some ocean regions as a result of covariation of precipitation with other climate processes, notably in the tropical Pacific (Figure 11b). The region of high precipitation in the western tropical Pacific (Figure 5b) migrates eastward during El Niño as ocean upwelling declines in the central and eastern tropical Pacific and increases in the western tropical Pacific [Philander, 1990]. Precipitation exerts only indirect control on ocean-atmosphere CO2 flux, via sea surface salinity (SSS). If precipitation exceeds evaporation (E–P < 0), SSS decreases and all chemical species in the surface ocean are diluted, leading to a lower concentration of dissolved inorganic carbon (therefore lower pCO2) but also lower total alkalinity (therefore lower CO2 solubility and higher pCO2); in general, the DIC effect dominates and excess precipitation will result in reduced pCO2 and ocean uptake.

[35] The (pointwise) correlation coefficients of ocean-atmosphere CO2 flux with SST, SSS, maximum winter mixed layer depth, and export production are shown in Figure 12. CO2 flux is negatively correlated with SST in the tropical ocean outgassing regions (cooler temperatures are associated with upwelling of carbon-rich water and increased outgassing) and positively correlated in the subtropics (cooler temperatures are associated with increased solubility and increased ocean uptake) (Figure 12a). CO2 flux is also positively correlated with SST in the subantarctic region of strong ocean uptake and negatively correlated in the region of high-latitude Southern Ocean outgassing (Figures 6b and 12a).

Figure 12.

Correlation coefficient (rxy) of ocean-atmosphere CO2 (positive upward) flux with sea surface temperature, sea surface salinity, maximum winter mixed layer depth, and export production. Export production is defined as organic carbon sedimentation across 100 m. Based on annual means except for mixing depth.

[36] CO2 flux is negatively correlated with SSS in the eastern North Pacific and positively correlated in the western tropical Pacific (Figure 12b). The western tropical Pacific and the eastern subarctic Pacific are both regions of excess precipitation and low salinity. In the western tropical Pacific, high salinity is associated with the eastward migration of convective precipitation during the El Niño–Southern Oscillation (ENSO) warm phase [Philander, 1990], which is in turn associated with ocean upwelling and a shallower thermocline in the western tropical Pacific [Bray et al., 1996; Merrifield et al., 1999], so salinity in this case may be a proxy for other factors (e.g., upwelling of DIC). The influence of DIC concentration/dilution on CO2 flux (positive correlation with salinity) is seen in all three low-latitude oceans, in most of the high-latitude Southern Ocean, and in parts of the high-latitude North Pacific and North Atlantic (Figure 12b). The spatial pattern is generally similar to that obtained by Doney et al. [2006] but has more east–west asymmetry in the tropical Pacific and slightly larger regions of negative correlation overall.

[37] In the study by Doney et al. [2006], as in the present study, there is a large contiguous region in the eastern North Pacific where SSS is negatively correlated with CO2 flux (Figure 12b), which is of opposite sign to what is expected from CO2 system thermodynamics if concentration/dilution of salt, DIC, and alkalinity was the underlying mechanism. In our model, negative correlation of SSS and CO2 flux is widespread in the midlatitudes (Figure 12b). The eastern subarctic Pacific is a region of excess precipitation and low salinity, as well as strong density stratification, and low biological production despite a generally upwelling favorable wind pattern [e.g., Gargett, 1991]. Lower salinity is associated with increased stratification and therefore reduced mixing and upwelling of DIC but also with reduced biological uptake. CO2 flux is negatively correlated with both export production and mixed layer depth (Figures 12c and 12d), i.e., there is greater uptake by the ocean when there is larger entrainment of nutrients and DIC by winter mixing. A negative correlation between CO2 flux and export production is observed over large areas of extratropical ocean (Figure 12d), indicating a strong biological effect on surface CO2 flux. Ocean CO2 uptake is associated with both higher export production and greater winter mixing depth, which implies a degree of biological control of CO2 exchange that is counter to established paradigms. Winter mixing should, in principle, bring more DIC to the surface than can be removed by the biota given the amount of excess nutrient entrained along with the DIC [e.g., Wagener and De Luca Rebello, 1987; Smith and Mackenzie, 1991]. However, excess alkalinity in deeper waters can produce exceptions to this rule; the mechanism is discussed at length in section 4.1.

[38] Biological control is also apparent in the Southern Ocean, where CO2 flux and export production are negatively correlated over large areas between 30°S and 60°S (Figure 12d). By contrast, in the model of Doney et al. [2006], there is a positive correlation between CO2 flux and particle export over most of the Southern Ocean. Our ocean carbon module CMOC has a high rate of organic export in the Southern Ocean (about twice the total for the North Atlantic and North Pacific combined), and modeled export is consistent with observation-based estimates [Zahariev et al., 2008].

3.5. Climate Modes

[39] The correlation of CO2 flux anomaly (F*) at each model grid point with time series of major modes of climate variability are shown in Figure 13. The NINO3 index is the mean SST anomaly for the region 5°S to 5°N, 90°W–150°W. The Pacific Decadal Oscillation (PDO) Index is the principal component of the first empirical orthogonal function of SST anomaly in the North Pacific (20°N–70°N) [Mantua et al., 1997]. The Southern Annular Mode (SAM) is calculated as the difference between zonal mean sea level pressure (SLP) at 40°S and 65°S [Gong and Wang, 1999]. An alternate definition based on the first principal component of SLP anomaly south of 20°S yields essentially the same answer (the two time series are correlated at r = 0.98). CO2 fluxes on land are largely uncorrelated with the indices, with the NINO3 index showing the strongest correlation (r = 0.17) (Table 5). However, use of annual mean data limits the utility of this analysis. The terrestrial contribution to the “ENSO-associated” CO2 flux is larger when seasonal lags are considered (Table 6); this is consistent with observation-based analyses [Jones and Cox, 2005].

Figure 13.

Correlation coefficient of surface CO2 flux (positive upward) with (a) the NINO3 index, (b) the Pacific Decadal Oscillation (PDO) index, and (c) the Southern Annular Mode (SAM) index. The NINO3 index is the mean SSTA for the region 5°S to 5°N, 90°W–150°W. The PDO index is the first principal component of sea surface temperature anomaly (SSTA) for the north Pacific (20°N–70°N). The SAM index is the difference between zonal mean sea level pressure at 40°S and 65°S.

Table 5. Correlation of Aggregate CO2 Flux (Total, Land Only, Ocean Only) with the NINO3 Index, the PDO Index, and the SAM Indexa
 NINO3PDOSAM
Total0.0870.0160.093
Land0.170.043−0.0023
Ocean−0.47−0.150.49
Table 6. Correlation of Total Tropical (≤20°) Land and Ocean CO2 Flux (FL, FO) and Tropical Terrestrial Precipitation with Mean SST Anomaly in the NINO3 Region (90°W–150°W, 5°N to 5°S), with SST Anomaly Leading by up to a Yeara
Lag (Months)Precipitation (Land)Land CO2 FluxOcean CO2 Flux
  • a

    Calculations are for 200 years of monthly model output with the mean seasonal cycle removed and a low-pass filter (25-point running mean) applied.

0−0.290.37−0.72
1−0.280.37−0.71
2−0.270.38−0.70
3−0.260.37−0.68
4−0.250.37−0.66
5−0.230.36−0.63
6−0.220.35−0.60
7−0.200.33−0.57
8−0.190.32−0.54
9−0.170.30−0.50
10−0.160.28−0.46
11−0.140.26−0.41
12−0.130.23−0.37

[40] Strong correlation of CO2 flux with the NINO3 index is largely localized to the tropical Pacific Ocean. The negative correlation in the equatorial zone is consistent with dynamical control of both CO2 flux and temperature, i.e., greater ocean outgassing is associated with upwelling of cold, DIC-rich subsurface water. The north and south Pacific subtropical gyres show a correlation with the NINO3 index, with a sign reversal in each hemisphere between the “transition zone” between the equatorial zone and the gyre (positive correlation, ∼5°–25° latitude) and the gyre center (negative correlation centered on ∼30° latitude) (Figure 13a). In the transition zone, the correlation is positive, opposite to the equatorial upwelling, which is consistent with increased uptake when the water being advected poleward out of the equatorial zone is cooler and more nutrient-enriched. But over much of the subtropics, the correlation is opposite in sign to the correlation with SST locally (Figure 12a). In other words, the primarily solubility-controlled subtropical oceans take up CO2 from the atmosphere when the local SST is cooler but take up less when equatorial Pacific SST is cooler. ENSO effects in subtropical gyres are subtle and difficult to detect but can include both decreased export production and increased storage of carbon in the form of dissolved organic carbon (DOC), which is not available for exchange with the atmosphere, as well as changes in temperature and wind variability with little or no change in the mean SST or wind speed [Karl et al., 1995]. A negative correlation with the NINO3 index (increased ocean uptake during warm ENSO) is consistent with the observations of Karl et al. [1995] that production of organic carbon increased during the 1991–1992 El Niño despite a reduction in export, with the excess accumulating as DOC. Globally, the correlation with the NINO3 index explains about a quarter of the variance in air-sea CO2 flux (Table 5).

[41] The PDO pattern is similar to the dipole pattern of the PDO mode of SST variability (warmer than usual temperatures in the tropical and eastern Pacific coincide with colder temperatures in the western subtropical and subarctic region), except that there is a sign reversal in the core of the equatorial upwelling zone (Figure 13b). The dipole is apparent between the northeastern Pacific (Gulf of Alaska) and the western Pacific (Kuroshio extension), where warmer temperatures lead to enhanced outgassing during positive and negative PDO phases, respectively. The region of negative correlation in the Gulf of Alaska extends throughout much the eastern subtropics, although it is weak or absent near the west coast of North America (Figure 13b). In the eastern equatorial Pacific, the competing effects of ocean dynamics (upwelling) and CO2-system thermodynamics (solubility) are shown by the reversal of the sign of the correlation within a region where SST anomalies are of the same sign in the PDO pattern. We explain this by noting that a positive SST anomaly in the equatorial zone implies weak upwelling and a deeper thermocline, so that reduced upwelling of DIC-rich subsurface water is the dominant effect (increased ocean uptake associated with warm SST during the positive PDO phase). Outside the equatorial upwelling zone, CO2 flux is primarily solubility controlled (decreased ocean uptake associated with warm SST during the positive PDO phase). There is a similar pattern of positive and negative correlations in the North Atlantic as the North Pacific, although the correlations are weaker (Figure 13b). The PDO explains substantially less of the total variance in air-sea CO2 flux than does the NINO3 or SAM indices (Table 5). For both PDO and ENSO, the effect on the terrestrial flux cancels out some of the ocean effect, and the fraction of variance of the total (land + ocean) flux explained by the PDO is very small (Table 5).

[42] CO2 fluxes in the southern midlatitudes and high latitudes are strongly associated with the SAM (Figure 13c), particularly in the region south of Australia where the strongest Ekman upwelling-associated outgassing occurs (Figure 6b). Here the wind anomalies associated with a positive SAM index are westerly and enhance equatorward Ekman transport and ocean outgassing [Hall and Visbeck, 2002]. The southern subtropics show a negative correlation with the SAM index over a broad area extending to quite low latitudes (Figure 13c). This was also seen in the calculations of Lovenduski et al. [2007, their Figure 7]. According to Hall and Visbeck [2002], a positive SAM implies anomalous easterly winds and poleward Ekman transport at these lower latitudes, which would tend to advect warmer water poleward, where it cools and absorbs CO2 from the atmosphere. The zonal distribution of this SAM-associated ocean sink is remarkably similar in our model and that of Lovenduski et al. [2007]: stronger in the western than the eastern South Pacific and weakest in the Atlantic (Figure 13c). The SAM explains about the same fraction of global air-sea CO2 flux as the NINO3 index (r = 0.49) (Table 5). Lovenduski et al. [2007] estimated a correlation of 0.69, considering only fluxes south of 35°S. The SAM has negligible impact on terrestrial fluxes (Table 5).

4. Discussion

4.1. Biological Control of Ocean Carbon Flux

[43] Over significant areas of extratropical ocean, ocean-atmosphere CO2 flux is negatively correlated with sea surface salinity, winter mixed layer depth, and export production (Figure 12). The model of Doney et al. [2006] produced a similar pattern (their Figure 13). In the subarctic Pacific, for example, a negative correlation of CO2 flux (positive upward) with SSS and mixed layer depth implies that upwelling of subsurface water or deeper winter mixing causes increased ocean uptake of CO2. This seems to contradict the established concept that ocean upwelling should result in outgassing because the “excess” DIC brought to the surface is at least equal to the amount of DIC that can be consumed by the biota given the amount of nutrient in the upwelled water [Smith and Mackenzie, 1991]. But there are also temperature, salinity, and alkalinity effects on solubility that in some regions combine to permit net uptake of atmospheric CO2. Temperature almost universally decreases with increasing depth, and alkalinity increases; in the subarctic Pacific, salinity also increases with depth. Upwelling thus brings to the surface colder, saltier water with elevated alkalinity, which limits conversion of HCO3 ions to CO2 and increases the solubility of CO2. So if the biota are able to consume the “Redfield equivalent” excess DIC, it is possible for upwelling or deep mixing to cause a decline in surface pCO2. This was also noted by Dutreuil et al. [2009] in a study of artificially induced vertical mixing.

[44] To assess this mechanism over a wider spatial scale using observations, we calculated pCO2 at the ocean surface for water from 200 m depth from global gridded data sets [Key et al., 2004; Antonov et al., 2006; Garcia et al., 2006; Locarnini et al., 2006]. Annual mean data were used, and the potential pCO2 of a 200 m water parcel was calculated at sea surface temperature and salinity, with the “Redfield equivalent” DIC of the excess (over the surface concentration) nitrate at 200 m removed. The calculation was performed for alkalinity at 200 m and for alkalinity at the surface (Figure 14). There are significant areas of subpolar ocean (poleward of 35°) where these water parcels would be undersaturated if transported instantaneously to the surface (Figure 14). The regions do not exactly overlap the regions that show negative correlations between CO2 flux and mixed layer depth in Figure 12c, but there is a general geographic correspondence. The excess (over surface values) alkalinity is a key factor without which the area of undersaturation is strongly attenuated, even with 100% utilization of excess nitrate, particularly in the North Pacific (Figure 14). The area of undersaturation is probably underestimated because we used sea surface temperature in the calculation, and deeper mixing would tend to cool the surface.

Figure 14.

pCO2 at the ocean surface calculated for 200 m water from NODC and GLODAP gridded data, calculated for (a) alkalinity at 200 m and (b) surface alkalinity. DIC is corrected for Redfield-equivalent excess nitrate (200 m nitrate concentration less surface concentration, multiplied by a C:N ratio of 6.625). Sea surface temperature and salinity were used in carbon chemistry and solubility calculations. Horizontal line indicates 288 ppm.

[45] Planktonic calcification in CMOC is strongly temperature dependent and, therefore, declines with increasing latitude [Zahariev et al., 2008]. Net uptake of atmospheric CO2 by recently upwelled water might not occur in regions with high rates of calcification because calcification removes alkalinity along with DIC, generating higher pCO2 despite a net drawdown of DIC. Extratropical uptake could be enhanced by reduced calcification due to ocean acidification, causing a negative feedback on atmospheric CO2 growth [Zondervan et al., 2001]. Our calculations also assume 100% utilization of nitrate, and CMOC erroneously simulates near-total drawdown of surface nitrate in the northeast Pacific [Zahariev et al., 2008; see also Gnanadesikan et al., 2004]. This large natural CO2 sink is therefore potentially sensitive to both ocean acidification and anthropogenically induced changes in iron supply from aeolian mineral dust [e.g., Elliott et al., 1997; Ridgwell et al., 2002; Mahowald and Luo, 2003; McConnell et al., 2007].

4.2. Interhemispheric Transport and Southern Ocean Outgassing

[46] Observation-based estimates of preindustrial interhemispheric (north–south) carbon transport by the ocean range from 0.2 to 0.8 Pg C yr−1, with the most likely values in the 0.3–0.35 range [Keeling and Peng, 1995; Mikaloff Fletcher et al., 2007]. Our model mean value is 0.38 Pg C yr−1, in contrast to some earlier models that had little or no interhemispheric transport [e.g., Murnane et al., 1999; Aumont et al., 2001]. Murnane et al. [1999] estimated a southward flux of 0.24 Pg C yr−1 in the Atlantic, but essentially all of this was cancelled out by northward fluxes in the Pacific and Indian Oceans. Aumont et al. [2001] estimated 0.35 Pg C yr−1 for a model that includes fluxes of fluvial C from the land but only 0.1 Pg C yr−1 for a model that does not include this term.

[47] In our model, the high-latitude (>50°S) Southern Ocean is a larger CO2 source (1.29 ± 0.11 Pg C yr−1) to the atmosphere than the Equatorial Pacific (0.76 ± 0.091 Pg C yr−1), but the Southern Ocean contribution is largely balanced by uptake within the midlatitude Southern Ocean (Figures 6 and 7). The Southern Ocean south of 30°S is nearly neutral with a small tendency toward net uptake (0.059 ± 0.13 Pg C yr−1). Uptake by the northern midlatitudes largely balances outgassing in the equatorial Pacific, with 0.46 ± 0.033 Pg C yr−1 taken up by the North Pacific and 0.30 ± 0.026 Pg C yr−1 by the North Atlantic. The standard deviations given are for annual data, so the interannual variability of the Southern Ocean is as large as that of the Equatorial Pacific, while the northern midlatitudes are much less variable. Lovenduski et al. [2007] estimated an interannual variability of Southern Ocean CO2 flux of ±0.19 Pg C yr−1, about 50% larger than ours.

[48] Jacobson et al. [2007] estimated that the preindustrial Southern Ocean outgassed at a rate of around 0.5 Pg C yr−1, using an inversion method that corrected ocean DIC data for the anthropogenic fraction, whereas they estimated values near zero with a forward model. They attributed this latter result to a sluggish meridional overturning and underestimation of interhemispheric transport. In our model, the North Atlantic meridional overturning streamfunction is ∼14 Sv (vs observation-based estimates of 15–18, see Zahariev et al. [2008] for sources), and the interhemispheric transport is 0.38 Pg C yr−1. So despite the somewhat weak meridional overturning in our model, the interhemispheric transport is consistent with observation-based estimates [e.g., Keeling and Peng, 1995; Mikaloff Fletcher et al., 2007], and high-latitude Southern Ocean outgassing is quite strong (Figures 6, 7, and 9). However, the Southern Ocean as a whole is a small net sink in our model (although the mean uptake is less than the year-to-year variability). We therefore speculate that weak meridional overturning may not entirely explain the absence of Southern Ocean outgassing in the model of Jacobson et al. [2007] and that their inversion results showing a net source in the preindustrial Southern Ocean may not be robust to the choice of regions and other assumptions made.

[49] Whether or not the Southern Ocean as a whole was a source of CO2 to the preindustrial atmosphere, these results emphasize the magnitude of the high-latitude Southern Ocean source in the preindustrial ocean, which provides additional context for evaluating hypotheses about glacial-interglacial changes in atmospheric CO2. Our source and sink regions roughly correspond to the “biogeochemical divide” identified by Marinov et al. [2006], who showed that increases in export production can have very different effects on atmospheric CO2, depending on which side of the divide they occur on. In their model, an enhanced biological pump has a greater effect on atmospheric CO2 in the high-latitude than the midlatitude Southern Ocean. In the context of our model of the preindustrial carbon cycle, this implies that glacial atmospheric CO2 may have responded not to an enhanced Southern Ocean sink but to an attenuated Southern Ocean source. This distinction may be particularly important in evaluating the role of proposed mechanisms that involve the equatorward expansion of the marginal ice zone around Antarctica [Stephens and Keeling, 2000; Gildor et al., 2002].

5. Summary and Conclusions

[50] The carbon cycle in the preindustrial control run of a new coupled climate/carbon-cycle model is described in detail. Major features include greater interannual variability of terrestrial relative to ocean CO2 exchange; net transport from the northern to the southern hemisphere in the ocean, maintaining an interhemispheric gradient in atmospheric CO2 that is the reverse of the present, anthropogenically dominated distribution; convective vertical transport of surface-generated tropical xCO2 anomalies and poleward horizontal advection of these near the tropopause; and high-latitude Southern Ocean outgassing accounting for most of the uptake by the midlatitude Southern Ocean sink. Despite the strength of the high-latitude Southern Ocean source, we find the Southern Ocean as a whole to be a small net sink for atmospheric CO2. Interhemispheric transport is estimated as 0.38 Pg C annually (southward in the ocean, northward in the atmosphere). Midlatitude to high-latitude oceans appear to show greater CO2 uptake in years with deeper winter mixing, which we attribute to excess alkalinity in subsurface waters and an efficient biological pump for removing excess DIC. This phenomenon appears to be widespread in the extratropical oceans in the model simulation and is consistent with offline calculations from observations. Terrestrial CO2 fluxes are controlled primarily by water availability at low latitudes, with temperature becoming increasingly important at higher latitudes. Physical controls on NEP are difficult to quantify because they represent the sum of the opposing effects on net primary production and heterotrophic respiration, both of which generally increase with increasing temperature and precipitation. However, physical controls on net primary production appear to be the most important influence on CO2 exchange because the amplitude of the response to changing environmental conditions is generally larger than for heterotrophic respiration. There are anomalies in both land and ocean flux associated with tropical Pacific sea surface temperature; the land component is significantly larger when seasonal lags are taken into account. The spatial patterns and variability of carbon fluxes in our model, and our inferences about physical controls, are largely consistent with observation-based and previous model studies [e.g., Jones and Cox, 2005; Zeng et al., 2005; Doney et al., 2006; Denman et al., 2007; Reichstein et al., 2007]. The model has a stable and realistic preindustrial mean climate and a reasonable amount of internal variability and so is useful as a baseline for historical CO2 emission and future emission scenario simulations.

Acknowledgments

[51] This research was funded in part by support from the Canadian Foundation for Climate and Atmospheric Sciences (network grant to K. Denman and N. Roulet). Nathan Gillett, Oleg Saenko, Knut von Salzen, and two anonymous reviewers made useful comments on an earlier draft.

Ancillary